Skip to main content
NCBI home page
NIHPA Author Manuscripts logoLink to NIHPA Author Manuscripts
. Author manuscript; available in PMC: 2018 Feb 2.
Published in final edited form as: J Biomech. 2016 Feb 10;49(5):718–725. doi: 10.1016/j.jbiomech.2016.02.006

Learning new gait patterns: Exploratory muscle activity during motor learning is not predicted by motor modules

Rajiv Ranganathan 1,2,3, Chandramouli Krishnan 1,4, Yasin Y Dhaher 1,2, William Z Rymer 1,2
PMCID: PMC5796520  NIHMSID: NIHMS871208  PMID: 26916510

Abstract

The motor module hypothesis in motor control proposes that the nervous system can simplify the problem of controlling a large number of muscles in human movement by grouping muscles into a smaller number of modules. Here, we tested one prediction of the modular organization hypothesis by examining whether there is preferential exploration along these motor modules during the learning of a new gait pattern. Healthy college-aged participants learned a new gait pattern which required increased hip and knee flexion during the swing phase while walking in a lower-extremity robot (Lokomat). The new gait pattern was displayed as a foot trajectory in the sagittal plane and participants attempted to match their foot trajectory to this template. We recorded EMG from 8 lower-extremity muscles and we extracted motor modules during both baseline walking and target-tracking using non-negative matrix factorization (NMF). Results showed increased trajectory variability in the first block of learning, indicating that participants were engaged in exploratory behavior. Critically, when we examined the muscle activity during this exploratory phase, we found that the composition of motor modules changed significantly within the first few strides of attempting the new gait pattern. The lack of persistence of the motor modules under even short time scales suggests that motor modules extracted during locomotion may be more indicative of correlated muscle activity induced by the task constraints of walking, rather than reflecting a modular control strategy.

Introduction

The degrees-of-freedom problem (Bernstein, 1967) is one of the long-standing problems in motor control – i.e., how does the motor system coordinate and control the numerous degrees of freedom in the body to produce goal-directed movement? One hypothesis is that the motor system groups muscles into functional units or motor modules and then controls each of these modules independently- cf. coordinative structures – (Easton, 1972; Kugler et al., 1980). In principle, this hierarchical organization could simplify the control problem since there is a reduction of the available degrees of freedom – i.e., instead of controlling the activity of each muscle independently, the nervous system only needs to control the activity of a smaller number of modules, each of which in turn regulates the activity of muscles within that module.

There have been several lines of evidence for the existence of motor modules in motor behavior. Initial approaches used simple pairwise correlation methods to infer that muscle activity of multiple muscles were correlated during tasks (Maier and Hepp-Reymond, 1995). More recent methods have used dimensionality reduction methods like PCA (Krishnamoorthy et al., 2003), factor analysis (Ivanenko et al., 2004) or non-matrix factorization (NMF) (d’Avella et al., 2003; Tresch et al., 1999), in which a single muscle may be part of more than one module. Evidence of motor modules has been provided in a wide repertoire of behaviors including balance (Ting and Macpherson, 2005; Torres-Oviedo and Ting, 2010), reaching (Cheung et al., 2009), isometric force production (Roh et al., 2012), and locomotion (Clark et al., 2010), with some of these papers providing evidence that that the composition of these motor modules is affected by neurological conditions such as stroke.

However, there has also been some debate about the evidence for such motor modules (Kutch and Valero-Cuevas, 2012). A central argument against inferring motor modules solely from statistical dimensionality reduction methods comes from examining the source of the correlations in the muscle activity – i.e. do the high correlations in activity seen in muscles within a module represent a neural control strategy or are they simply a by-product of simultaneous muscle activity due to task and biomechanical (i.e. “non-neural”) constraints (Buchanan et al., 1986; Kutch and Valero-Cuevas, 2011)? Several modeling studies have shown that despite the large number of muscles present, the musculoskeletal system is not redundant for many tasks and therefore biomechanical constraints may lead to correlated muscle activity in certain portions of the workspace (Kutch and Valero-Cuevas, 2011). Furthermore, studies that have examined correlations in the variability of the muscle activation (which is less influenced by these task constraints when compared to the average muscle activation) have shown little support for the motor modules hypothesis (Ranganathan and Krishnan, 2012; Valero-Cuevas et al., 2009). One methodological caveat however is that variability in EMG is affected by low signal-to-noise ratios and therefore, it still remains unclear whether motor modules extracted by dimensionality reduction methods truly reflect a neural control strategy (see Figure 1).

Figure 1.

Figure 1

Open in a new tab

Schematic of alternative models of correlated muscle activity. On the left panel (A), there is a common input to both muscles M1 and M2. On the right panel (B), the two muscles share independent inputs that are correlated. Dimensionality reduction methods in these two cases will yield identical muscle modes shown (central panel, top). However, these two conditions can be differentiated when learning a novel task – the EMG activity of the two muscles will still share the same relation in (case A), but may alter their relation in (case B) (central panel, top).

Given that there are several versions of the motor module hypothesis that either keep muscle activation patterns (Ivanenko et al., 2004) or muscle weightings invariant (Ting and Macpherson, 2005), our focus was restricted to the version where the muscle weightings in each motor module were invariant. Here, we tested one prediction of this particular hypothesis – i.e., if motor modules truly reflect a neural control strategy during a task, then there should be preferential exploration along these modules when attempting to learn a new variation of the task – i.e., the muscle activity during the initial phases of learning should be explained largely with the same motor modules as that observed during the original task. Previous studies have shown that the initial strategies that participants use to adapt to new tasks are heavily influenced by both habitual (de Rugy et al., 2012) as well as previously learned coordination strategies (Kobak and Mehring, 2012; Ranganathan et al., 2014). To test this prediction, we used a locomotion task where the novel task was to modify the kinematics during swing phase of locomotion. We specifically focused the manipulation on the swing phase and the associated hamstring activity because it is less constrained mechanically (since the foot is not in contact with the ground and has minimal interaction with the environment) and so there are different ways to change the hamstring muscle activation during the swing phase while still being able to walk. The hypothesis was that if motor modules are a neural control strategy, then the motor modules during initial phases of learning the novel gait pattern should resemble the motor modules during normal walking.

Methods

Participants

Participants (n = 7) were healthy young males (age range 18–35 years) who were free of neurological or musculoskeletal injury. All participants were right-leg dominant as determined by their preferred leg for kicking a ball (Krishnan and Williams, 2009). Written informed consent was obtained from each participant and all procedures were approved by the Institutional Review Board at Northwestern University.

Equipment

Participants walked in a lower-extremity robotic exoskeleton (Lokomat, Hocoma, Switzerland). The Lokomat was configured to be in a “co-operative control mode”, which essentially means that the participant had to actively walk in order to produce a walking pattern (i.e., the robot did not move the participant’s legs passively) (Duschau-Wicke et al., 2010; Krishnan et al., 2013a). In addition, the stiffness of the robot was minimal in this mode so that participant could exert force against the robot and change the locomotion pattern if required. There was also a computer monitor in front of the participant which displayed the x-y position of the lateral malleolus in the sagittal plane (hereafter referred to as the “foot trajectory”). The foot trajectory was computed using a forward-kinematic model that uses the hip and knee joint angles recorded from the Lokomat (sampling rate = 1000 Hz) and also the segment lengths of the thigh and shank (Krishnan et al., 2012).

Surface EMG signals (Motion Labs, Baton Rouge, LA) from 8 muscles on the right leg were recorded simultaneously. The muscles recorded were vastus medialis (VM), rectus femoris (RF), medial hamstring (MH), lateral hamstring (LH), tibialis anterior (TA), medial gastrocnemius (MG), soleus (SO), and gluteus medius (GM). The placement of the electrodes conformed to the SENIAM guidelines (www.seniam.org) except for SO, for which the electrode was placed at 2/3rd of the line starting from the lateral femoral condyle to the lateral malleolus (Ranganathan and Krishnan, 2012)

Protocol

The schematic of the experimental protocol is given in Figure 2. Participants were strapped to the Lokomat using Velcro cuffs at the pelvis, thigh and shank and the robot adjusted so that the hip and knee joints of the exoskeleton and the participant were aligned.

Figure 2.

Figure 2

Open in a new tab

(A) Schematic of experimental setup in the robotic exoskeleton. Participants viewed a target foot-trajectory on the screen and attempted to match their own foot-trajectory with the target. (B, top panel) Construction of the target-template was achieved by increasing the hip and knee angle by 20% during the swing phase (see Methods). (B, bottom panel). Schematic of computation of tracking error – tracking error was computed as the non-overlapping area between the target foot trajectory and the actual foot trajectory produced in that gait cycle (shaded region). The resulting area (i.e., the error) was normalized to the area within the participant’s target foot trajectory.

First, participants were given a 5-minute familiarization phase to get them used to walking in the Lokomat. After this familiarization phase, participants walked inside the Lokomat with no visual feedback for 2 minutes at 2 km/hr. At the end of this phase, we computed the average hip and knee angles across the gait cycle (averaged over multiple cycles) when walking inside the Lokomat. We then computed the target foot trajectory by increasing the average hip and knee angles by 20% during the swing phase. To prevent an abrupt change of 20% both in the beginning and at the end of the swing phase, we smoothed the trajectory using a Hanning window.

Once the target foot trajectory was computed, it was displayed on the computer monitor in front of the participant. In addition, the participant could also see concurrent visual feedback of his actual foot trajectory while walking in the Lokomat. The duration of the concurrent visual feedback on the screen was also adjusted so that, in addition to the instantaneous feedback of the foot trajectory, the participant could see the entire trajectory that he had produced over the last gait cycle.

After the target was computed and displayed, participants performed seven 2-minute blocks of target-tracking (where they received visual feedback and were instructed to change their gait to match the new target) alternated with seven blocks of baseline walking (where they were instructed to just walk normally and received no visual feedback).

Data Analysis

Tracking error and Variability

In the target-tracking blocks, the tracking error for each gait cycle was computed as the non-overlapping area (in pixels) between the target foot trajectory and the actual foot trajectory produced in that gait cycle (Figure 2B). This area was normalized to area within the participant’s target foot trajectory. This metric quantified how the actual trajectory produced varied from the target trajectory.

For computing variability in the gait pattern, we computed the non-overlapping area between the ensemble average foot trajectory and the actual foot trajectory produced in that gait cycle. This area was normalized to the area within the participant’s target foot trajectory. This metric quantified how the actual trajectory produced varied from the mean trajectory.

Motor Modules

In order to extract motor modules, we used non-negative matrix factorization (NMF) that has been used in several other studies (Clark et al., 2010). EMG data were first high-pass filtered at 20 Hz, rectified and low-pass filtered at 6 Hz using a zero phase-lag Butterworth filter (8th order). EMGs in individual muscles were normalized to the maximum ensemble-averaged muscle activity in the first baseline walking trial.

The muscle activity matrix M was of size N × 8 where N represents the number of data points and 8 represents the number of muscles. NMF decomposes the muscle activity M into two matrices – one which represents a time-invariant composition of the motor modules (in terms of weighting of the different muscles), and the other represents the time-varying activity of these modules (Tresch et al., 2006).

Preferential exploration

In order to compare if there was preferential exploration along the existing motor modules when learning the novel task, we used the principal angle (Cheung et al., 2009) to compare the similarity of modules during baseline walking and during the first target-tracking block, and between the baseline walking and the final target-tracking block. The principal angles were computed in MATLAB (Mathworks, Natick MA) and we computed the average of the four principal angles obtained. This average principal angle measures the similarity between the motor modules, with higher angles indicating that the modules are more dissimilar (Todorov and Ghahramani, 2004).

Statistical Analysis

To examine the initial adaptation to the target-tracking task, we used a paired t-test to compare the last block of baseline walking against the first block of target-tracking. To examine the effect of practice, we used paired t-tests to compare the error in the first and last block of target-tracking. To examine the changes in EMG activity of the hamstring muscles, we used a one-way repeated measures analysis of variance, with block (baseline, first block, and final block of target-tracking) as a within-subjects factor. Finally, to test our main hypothesis about the change in motor modules using the average principal angle, we ran a bootstrapping procedure where we randomly selected 50% of the strides during baseline walking and computed the average principal angle of the motor modules extracted from this data with: (i) the motor modules extracted from the remaining strides during baseline walking, (ii) the motor modules extracted from 50% of the strides during the first target-tracking block, and (iii) the motor modules extracted from 50% of the strides during the last target-tracking block. We did this for 50 iterations and used a t-test (on each individual participant) to compare if the motor modules during the target-tracking blocks were significantly different from the motor modules during baseline walking.

Results

Tracking error and Variability

Sample foot-trajectories of a typical participant during the different phases of the experiment are shown in Figure 3A. The variability in the gait pattern in the first block of practice was significantly higher than the variability during baseline walking (t(6) = 2.560, p = 0.043), suggesting that this was a period of exploration by the participant in attempting to generate the novel gait pattern. In addition, tracking error decreased with practice, suggesting that participants were able to modify their gait pattern to adapt their gait to match the novel gait pattern (t (6) = 2.745, p = 0.034) (Figure 3B).

Figure 3.

Figure 3

Open in a new tab

(A) Sample foot trajectories of a participant during baseline (i.e., when there was no target), and during early and late practice of the novel coordination pattern. (B). Tracking variability and tracking error as a function of practice of the novel gait task. Error bars represent between-participant standard error.

Change in EMG activity in the hamstring muscles

We also compared the EMG activity in the targeted muscles (i.e. the hamstring muscles) during baseline, the first block of target-tracking and the final block of target-tracking. As expected, our analysis showed statistically significant increases in both MH and LH muscle activity (Figure 4A), specifically during the initial part of the swing phase (MH: F(2,12) = 9.322, p = 0.004; LH: F(2,12) = 9.807, p = 0.003) (Figure 4B). In both muscles, the muscle activity during baseline was significantly different from the first and last block of target-tracking (p = 0.005 to 0.028). The muscle activity in the both muscles between the first and last block was not statistically significant (p = 0.136 and p = 0.372).

Figure 4.

Figure 4

Open in a new tab

(A) Mean ensemble averaged EMG (across all participants) of the medial and lateral hamstring muscles during baseline walking and after learning of the novel gait pattern. (B). Mean muscle activity for the same two muscles averaged during the swing phase. Error bands and error bars represent between-participant standard error.

Motor Modules

The analysis of the EMG activity during baseline walking using NMF revealed that four motor modules could account for over 85% of the overall variance in the data (Mean VAF = 88±2%). The composition of these motor modules is shown in Figure 5 and these modules were qualitatively similar to those reported in the literature when subjects simply walked on a treadmill (indicating that walking inside the Lokomat did not significantly change these modules).

Figure 5.

Figure 5

Open in a new tab

Motor module composition obtained during baseline walking (averaged across all participants). The four modules accounted for over 85% of the variance in the data. Error bars represent between-participant standard error.

Preferential exploration using Principal Angle

There was a significant increase in the average principal angle from baseline walking to the first target-tracking block, indicating that there was a change in the composition of the motor modules even when first attempting the novel gait pattern. This was significant for all 7 participants on an individual level (ps <0.01) with the average principal angle going from 6±3° (i.e. the variation during baseline walking) to 21±8° during the first target-tracking block. This difference between baseline walking and target-tracking persisted until the end of the training (i.e. the final target-tracking block), where the average principal angle with the baseline block was 22±5° (Figure 6).

Figure 6.

Figure 6

Open in a new tab

(A) Composition of the hamstrings motor module averaged across participants during the last 30 strides of baseline walking, the first 30 strides of target-tracking during the first block of practice, and the first 30 strides of target-tracking during the final block of practice. Error bars represent between-participant standard error. (B) Average principal angle between the motor modules from the baseline block and these three phases for each participant using a bootstrap procedure. Error bars represent standard error for each participant from the bootstrap analysis.

Discussion

Control strategies in locomotion have been examined at different levels in the neuromuscular system – at the level of the task (Dingwell et al., 2010), at the level of joint kinematics (Krishnan et al., 2013b) and at the level of muscles (Clark et al., 2010). The focus of the current study was at the muscular level and we tested one prediction of the motor module hypothesis – i.e., if the control of individual muscles in locomotion was achieved by controlling a smaller number of motor modules, then there should be preferential exploration along these modules when attempting to learn a new task. We created a condition where participants had to learn a novel gait pattern by changing the foot trajectory to match a target template. Participants were able to learn this task, as indicated by a reduction in their error and variability over several blocks of practice, and also showed associated changes in the hamstrings muscle activity. However, our results showed no evidence of any preferential exploration – we found motor modules during baseline walking and target-tracking were substantially different and these differences were observed almost immediately at the beginning of target-tracking. These results provide evidence against the hypothesis that the muscle activity during locomotion is regulated through motor modules where muscle weightings are invariant.

Other studies on learning have shown that muscle weightings are changed when learning a novel motor task (Kargo and Nitz, 2003). However, it is important to note we designed the task so that a change in motor modules was not necessitated by the task itself. Specifically, the task was designed so that the hamstring activity in the swing phase had to be changed. Therefore while an increase in hamstring muscle activity was expected to match the new target pattern, there was no mechanical requirement to alter the relative muscle activation of the medial and lateral hamstring muscles. One alternative explanation for the results is that the nervous system might use motor modules for normal walking but instantaneously switched to a new set of motor modules to explore the solutions required for achieving the new gait pattern. We see two potential problems with this explanation – first, at least some versions of the motor module hypothesis include discussion of neural substrates (Clark et al., 2010), and it is unlikely that these hard-wired constraints can change on such short time scales. Second, even if we assume that the motor modules are not hard-wired, but instead they are flexibly assembled for a particular task context, the inability for the motor modules to predict muscle activity even during the initial exploratory phase of learning a novel task raises the question of whether this reflects a control strategy or is instead possibly an artifact of task-induced correlations in muscle activity (Kutch and Valero-Cuevas, 2011; Ranganathan and Krishnan, 2012). Finally, although some have suggested that there is a vast repertoire of motor modules and that only a limited number are used for a given task (Chiel et al., 2009), this notion of motor modules becomes potentially untestable (de Rugy et al., 2013), especially if these new motor modules can be recruited in the relatively short time scales observed here.

We would like to emphasize that the current results do not rule out the presence of motor modules per se. There is evidence, specifically from the dynamical systems view of coordination that coordination patterns are not built from a blank slate and that there are certain preferred patterns which shape motor learning (Zanone and Kelso, 1992). Even though these studies have predominantly been done in the context of coordinating two effectors (e.g. bimanual coordination), Berger and colleagues showed in the context of multi-muscle coordination that it is possible to construct virtual tasks of different degrees of difficulty depending on whether they can be achieved using the existing coordination patterns (Berger et al., 2013). However, here the focus was on one specific version of the motor module hypothesis – which predicted that the weighting of each muscle within the module remains constant. This evidence is in line with other studies that have shown that the relative contribution of synergistic muscles changes as a function of mechanical demands on the limb (Howard et al., 1986; McGowan et al., 2010; Nelson and Roberts, 2008; Wakeling and Horn, 2009). One previous study showed that altering the mechanical demands (e.g. changing body mass/weight) tends to alter the muscle weightings within a motor module (McGowan et al., 2010), but these results were interpreted as “fine-tuning” of existing motor modules as the time scale of the change was not considered. Here, we extend this result by showing that not only are the modules different under different mechanical demands, the change in composition of the modules happens in rather short time scales during the initial exploration phase of the new gait pattern, and therefore the identified modules are unlikely to reflect a neural control strategy.

One limitation of our study is that the Lokomat restricted all motion to the sagittal plane. Therefore, it is possible that we did not adequately capture exploration across any new motor modules that would normally result in motion outside the sagittal plane (e.g. abduction/adduction). However, we believe that any exploration outside of the sagittal plane is unlikely since participants were familiarized with the Lokomat at the beginning of the experiment and were likely aware of how stiff the Lokomat is along these directions. Furthermore, the fact that all participants showed similar changes in motor modules between the baseline walking and the initial exploration phase, and that the motor modules were different from baseline even at the end of learning (where there would be minimal exploration) suggests that the changes observed are not confounded by recruitment of new motor modules. In addition, even though our motor modules were extracted from a single walking speed condition, the motor modules extracted were similar to those studies where modules were extracted from multiple walking speeds(see for e.g. Clark et al., 2010).

The present results definitely highlight the potential problem of relying only on statistical dimensionality reduction methods to extract motor modules (Burkholder and van Antwerp, 2013), especially in rhythmic activities like locomotion, where there is already a high degree of correlation between muscles simply as a consequence of task constraints (Ranganathan and Krishnan, 2012). Moreover, another argument that has been made against the statistical approach (where the primary goal is to maximize the variance accounted for) is that it does not take into account the non-linearities in the motor system – therefore even small errors in reconstructing the muscle activity may actually have large consequences at the behavioral level (de Rugy et al., 2013). Taken together, these results suggest that future tests of the hypothesis should focus on behavioral predictions of the motor module hypothesis under new conditions rather than simply relying on goodness of statistical fits (Steele et al., 2013; Tresch and Jarc, 2009).

Conclusion

In summary, we tested one version of the motor module hypothesis which assumes that the muscle weightings within a motor module are invariant. This hypothesis predicts that muscle activity recorded when performing a novel gait pattern should show preferential exploration along existing motor modules. However, we found no evidence for any such preferential exploration, and changes in motor modules were observed almost immediately at the beginning of learning the novel task. These results point out that motor modules in locomotion as extracted by dimensionality reduction methods may not be truly reflective of underlying neural control strategies.

Acknowledgments

Grant Support: This work was supported by a MARS-RERC grant H133E070013 funded by NIDRR, and by grants R03HD069806 and R01EB019834 from the National Institutes of Health. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript

References

  1. Berger DJ, Gentner R, Edmunds T, Pai DK, d’Avella A. Differences in adaptation rates after virtual surgeries provide direct evidence for modularity. Journal of Neuroscience. 2013;33:12384–12394. doi: 10.1523/JNEUROSCI.0122-13.2013. [DOI] [PMC free article] [PubMed] [Google Scholar]
  2. Bernstein NA. The co-ordination and regulation of movements. Pergamon Press; Oxford: 1967. [Google Scholar]
  3. Buchanan TS, Almdale DP, Lewis JL, Rymer WZ. Characteristics of synergic relations during isometric contractions of human elbow muscles. Journal of Neurophysiology. 1986;56:1225–1241. doi: 10.1152/jn.1986.56.5.1225. [DOI] [PubMed] [Google Scholar]
  4. Burkholder TJ, van Antwerp KW. Practical limits on muscle synergy identification by non-negative matrix factorization in systems with mechanical constraints. Medical and Biological Engineering and Computing. 2013;51:187–196. doi: 10.1007/s11517-012-0983-8. [DOI] [PMC free article] [PubMed] [Google Scholar]
  5. Cheung VC, Piron L, Agostini M, Silvoni S, Turolla A, Bizzi E. Stability of muscle synergies for voluntary actions after cortical stroke in humans. Proceedings of the National Academy of Sciences of the United States of America. 2009;106:19563–19568. doi: 10.1073/pnas.0910114106. [DOI] [PMC free article] [PubMed] [Google Scholar]
  6. Chiel HJ, Ting LH, Ekeberg O, Hartmann MJ. The brain in its body: motor control and sensing in a biomechanical context. Journal of Neuroscience. 2009;29:12807–12814. doi: 10.1523/JNEUROSCI.3338-09.2009. [DOI] [PMC free article] [PubMed] [Google Scholar]
  7. Clark DJ, Ting LH, Zajac FE, Neptune RR, Kautz SA. Merging of healthy motor modules predicts reduced locomotor performance and muscle coordination complexity post-stroke. Journal of Neurophysiology. 2010;103:844–857. doi: 10.1152/jn.00825.2009. [DOI] [PMC free article] [PubMed] [Google Scholar]
  8. d’Avella A, Saltiel P, Bizzi E. Combinations of muscle synergies in the construction of a natural motor behavior. Nature Neuroscience. 2003;6:300–308. doi: 10.1038/nn1010. [DOI] [PubMed] [Google Scholar]
  9. de Rugy A, Loeb GE, Carroll TJ. Muscle coordination is habitual rather than optimal. Journal of Neuroscience. 2012;32:7384–7391. doi: 10.1523/JNEUROSCI.5792-11.2012. [DOI] [PMC free article] [PubMed] [Google Scholar]
  10. de Rugy A, Loeb GE, Carroll TJ. Are muscle synergies useful for neural control? Frontiers in Computational Neuroscience. 2013;7:19. doi: 10.3389/fncom.2013.00019. [DOI] [PMC free article] [PubMed] [Google Scholar]
  11. Dingwell JB, John J, Cusumano JP. Do humans optimally exploit redundancy to control step variability in walking? PLoS Computational Biology. 2010;6:e1000856. doi: 10.1371/journal.pcbi.1000856. [DOI] [PMC free article] [PubMed] [Google Scholar]
  12. Duschau-Wicke A, von Zitzewitz J, Caprez A, Lunenburger L, Riener R. Path control: a method for patient-cooperative robot-aided gait rehabilitation. IEEE Transactions on Neural Systems and Rehabilitation Engineering. 2010;18:38–48. doi: 10.1109/TNSRE.2009.2033061. [DOI] [PubMed] [Google Scholar]
  13. Easton TA. On the normal use of reflexes. American Scientist. 1972;60:591–599. [PubMed] [Google Scholar]
  14. Howard JD, Hoit JD, Enoka RM, Hasan Z. Relative activation of two human elbow flexors under isometric conditions: a cautionary note concerning flexor equivalence. Experimental Brain Research. 1986;62:199–202. doi: 10.1007/BF00237416. [DOI] [PubMed] [Google Scholar]
  15. Ivanenko YP, Poppele RE, Lacquaniti F. Five basic muscle activation patterns account for muscle activity during human locomotion. Journal of Physiology. 2004;556:267–282. doi: 10.1113/jphysiol.2003.057174. [DOI] [PMC free article] [PubMed] [Google Scholar]
  16. Kargo WJ, Nitz DA. Early skill learning is expressed through selection and tuning of cortically represented muscle synergies. Journal of Neuroscience. 2003;23:11255–11269. doi: 10.1523/JNEUROSCI.23-35-11255.2003. [DOI] [PMC free article] [PubMed] [Google Scholar]
  17. Kobak D, Mehring C. Adaptation paths to novel motor tasks are shaped by prior structure learning. Journal of Neuroscience. 2012;32:9898–9908. doi: 10.1523/JNEUROSCI.0958-12.2012. [DOI] [PMC free article] [PubMed] [Google Scholar]
  18. Krishnamoorthy V, Goodman S, Zatsiorsky V, Latash ML. Muscle synergies during shifts of the center of pressure by standing persons: identification of muscle modes. Biological Cybernetics. 2003;89:152–161. doi: 10.1007/s00422-003-0419-5. [DOI] [PubMed] [Google Scholar]
  19. Krishnan C, Ranganathan R, Dhaher YY, Rymer WZ. A pilot study on the feasibility of robot-aided leg motor training to facilitate active participation. PLoS One. 2013a;8:e77370. doi: 10.1371/journal.pone.0077370. [DOI] [PMC free article] [PubMed] [Google Scholar]
  20. Krishnan C, Ranganathan R, Kantak SS, Dhaher YY, Rymer WZ. Active robotic training improves locomotor function in a stroke survivor. Journal of Neuroengineering and Rehabilitation. 2012;9:57. doi: 10.1186/1743-0003-9-57. [DOI] [PMC free article] [PubMed] [Google Scholar]
  21. Krishnan C, Williams GN. Variability in antagonist muscle activity and peak torque during isometric knee strength testing. Iowa Orthopaedic Journal. 2009;29:149–158. [PMC free article] [PubMed] [Google Scholar]
  22. Krishnan V, Rosenblatt NJ, Latash ML, Grabiner MD. The effects of age on stabilization of the mediolateral trajectory of the swing foot. Gait and Posture. 2013b;38:923–928. doi: 10.1016/j.gaitpost.2013.04.023. [DOI] [PubMed] [Google Scholar]
  23. Kugler PN, Kelso JAS, Turvey MT. On the concept of coordinative structures as dissipative structures: I. Theoretical lines of convergence. In: Stelmach GE, Requin J, editors. Tutorials in Motor Behavior. North-Holland; Amsterdam: 1980. pp. 3–47. [Google Scholar]
  24. Kutch JJ, Valero-Cuevas FJ. Muscle redundancy does not imply robustness to muscle dysfunction. Journal of Biomechanics. 2011;44:1264–1270. doi: 10.1016/j.jbiomech.2011.02.014. [DOI] [PMC free article] [PubMed] [Google Scholar]
  25. Kutch JJ, Valero-Cuevas FJ. Challenges and new approaches to proving the existence of muscle synergies of neural origin. PLoS Computational Biology. 2012;8:e1002434. doi: 10.1371/journal.pcbi.1002434. [DOI] [PMC free article] [PubMed] [Google Scholar]
  26. Maier MA, Hepp-Reymond MC. EMG activation patterns during force production in precision grip. II. Muscular synergies in the spatial and temporal domain. Experimental Brain Research. 1995;103:123–136. doi: 10.1007/BF00241970. [DOI] [PubMed] [Google Scholar]
  27. McGowan CP, Neptune RR, Clark DJ, Kautz SA. Modular control of human walking: Adaptations to altered mechanical demands. Journal of Biomechanics. 2010;43:412–419. doi: 10.1016/j.jbiomech.2009.10.009. [DOI] [PMC free article] [PubMed] [Google Scholar]
  28. Nelson FE, Roberts TJ. Task-dependent force sharing between muscle synergists during locomotion in turkeys. Journal of Experimental Biology. 2008;211:1211–1220. doi: 10.1242/jeb.013227. [DOI] [PubMed] [Google Scholar]
  29. Ranganathan R, Krishnan C. Extracting synergies in gait: using EMG variability to evaluate control strategies. Journal of Neurophysiology. 2012;108:1537–1544. doi: 10.1152/jn.01112.2011. [DOI] [PMC free article] [PubMed] [Google Scholar]
  30. Ranganathan R, Wieser J, Mosier KM, Mussa-Ivaldi FA, Scheidt RA. Learning redundant motor tasks with and without overlapping dimensions: facilitation and interference effects. Journal of Neuroscience. 2014;34:8289–8299. doi: 10.1523/JNEUROSCI.4455-13.2014. [DOI] [PMC free article] [PubMed] [Google Scholar]
  31. Roh J, Rymer WZ, Beer RF. Robustness of muscle synergies underlying three-dimensional force generation at the hand in healthy humans. Journal of Neurophysiology. 2012;107:2123–2142. doi: 10.1152/jn.00173.2011. [DOI] [PMC free article] [PubMed] [Google Scholar]
  32. Steele KM, Tresch MC, Perreault EJ. The number and choice of muscles impact the results of muscle synergy analyses. Frontiers in Computational Neuroscience. 2013;7:105. doi: 10.3389/fncom.2013.00105. [DOI] [PMC free article] [PubMed] [Google Scholar]
  33. Ting LH, Macpherson JM. A limited set of muscle synergies for force control during a postural task. Journal of Neurophysiology. 2005;93:609–613. doi: 10.1152/jn.00681.2004. [DOI] [PubMed] [Google Scholar]
  34. Todorov E, Ghahramani Z. Analysis of the synergies underlying complex hand manipulation. Conference Proceedings – IEEE Engineering in Medicine and Biology Society. 2004;6:4637–4640. doi: 10.1109/IEMBS.2004.1404285. [DOI] [PubMed] [Google Scholar]
  35. Torres-Oviedo G, Ting LH. Subject-specific muscle synergies in human balance control are consistent across different biomechanical contexts. Journal of Neurophysiology. 2010;103:3084–3098. doi: 10.1152/jn.00960.2009. [DOI] [PMC free article] [PubMed] [Google Scholar]
  36. Tresch MC, Cheung VC, d’Avella A. Matrix factorization algorithms for the identification of muscle synergies: evaluation on simulated and experimental data sets. Journal of Neurophysiology. 2006;95:2199–2212. doi: 10.1152/jn.00222.2005. [DOI] [PubMed] [Google Scholar]
  37. Tresch MC, Jarc A. The case for and against muscle synergies. Current Opinion in Neurobiology. 2009;19:601–607. doi: 10.1016/j.conb.2009.09.002. [DOI] [PMC free article] [PubMed] [Google Scholar]
  38. Tresch MC, Saltiel P, Bizzi E. The construction of movement by the spinal cord. Nature Neuroscience. 1999;2:162–167. doi: 10.1038/5721. [DOI] [PubMed] [Google Scholar]
  39. Valero-Cuevas FJ, Venkadesan M, Todorov E. Structured variability of muscle activations supports the minimal intervention principle of motor control. Journal of Neurophysiology. 2009;102:59–68. doi: 10.1152/jn.90324.2008. [DOI] [PMC free article] [PubMed] [Google Scholar]
  40. Wakeling JM, Horn T. Neuromechanics of muscle synergies during cycling. Journal of Neurophysiology. 2009;101:843–854. doi: 10.1152/jn.90679.2008. [DOI] [PubMed] [Google Scholar]
  41. Zanone PG, Kelso JA. Evolution of behavioral attractors with learning: nonequilibrium phase transitions. Journal of Experimental Psychology: Human Perception and Performance. 1992;18:403–421. doi: 10.1037//0096-1523.18.2.403. [DOI] [PubMed] [Google Scholar]

RESOURCES